PFP NFTs: not priced by the content of their character
Earlier last week, Jason Watkins (knocked.eth) posted a question that I found very thought-provoking:
Where in the developed world might you find a dollar value placed on skin tone?
The surprising answer he gave: OpenSea
Jason’s observation was that there are a number of personal profile picture NFT projects out there that feature jpgs of characters with different skin colors, and the floor prices for tokens with underlying artwork comprising darker skinned avatars appeared to be lower than those for lighter skinned avatars. (The floor price is the lowest-priced NFT within an NFT collection, and it is set by the seller of that NFT.)
The primary example he gave was the World of Women collection, which features 10,000 female avatars with different hair styles, accessories, clothing, and of course, skin tones.
Apparent racial bias in collectibles is nothing new — even a couple of decades ago a simple school science fair study in Boulder, CO by an eight year old concluded that the children of the school preferred white Barbie dolls to black Barbie dolls. The school subsequently removed the experiment from the fair, and various postmortems on the project raised a lot of important questions regarding freedom of speech and racial issues.
It sure looks that way
I browsed the Buy Now options on offer for the collection, and I had to admit, my first impression was that Jason was indeed right. It looked like darker skin tones were, on the whole, selling for less.
However, first impressions can be misleading, and it is easy to make a snap judgement, especially if you’re anticipating a certain outcome. OpenSea is a marketplace, so if the buyers on that marketplace as a whole hold discriminatory views, then you would expect those views to be reflected in the pricing for the assets on offer, but it is worth being at least reasonably thorough.
And so I took an hour to analyse the data on offer for the World of Women collection. I didn’t have the time to conduct a rigorous analysis, by considering data over a long period of time, or to write some code to automatically scrape the Ethereum blockchain and the OpenSea site, but I reasoned that anything must be better than a five second glance and an indignant sigh from my position of privilege before moving on to the next post in my feed.
Methodology and results
World of Women lists the different skin tone categories — fourteen in total — in the NFTs metadata as attributes, which are conveniently listed and selectable on the OpenSea collection page. However, things are complicated a bit by the fact that there are different numbers of tokens in each group.
I decided a simple spreadsheet would be the fasted and easiest tool to use for an initial test of the hypothesis, and so I created an table containing the skin tone description, the total number of tokens in each tone group, and the actual color. I determined the color value using a color picker tool by clicking between the eyes of a suitable avatar not wearing glasses.
Initially I added the floor price, median price, and maximum price, but then reconsidered. The minimum and maximum prices were outliers in some of the collections, and so I swapped them for the first price after the floor price, and the last price before the maximum. I know this is not terribly scientific, and with a bit more time someone could plot curves using all the prices on offer, calculate the arithmetic average, and perhaps run some statistical methods over the data sets, but I was doing this in my lunch hour so I had to chose something a little less labor intensive.
The next step was to sort the table three times — once for the the price immediately after the floor price, once for the median, and once for the price just before the maximum. Each time in ascending order.
Quatorze points
Then I used a modified version of the scoring system for the Eurovision song festival, by adding a points score for each row — one point for the first row, two points for the second, and so on up to 14 points for the last and most expensive row.
Again, I confess that using a song scoring scheme is not the most rigorous approach to all of this, but given that I didn’t know what the outcome was going to be, I think it’s reasonable for a “pilot study”. There are some darker colors in the lower parts of each of the tables, so it is not immediately obvious what the aggregate result is going to be.
Add it up
Finally, I created a single table in which the three scores for each color were summed, and then sorted the table from the lowest score (representing the least desirable tokens) to the highest.
From this we can deduce that “burning red” and “cyber green” tokens are the least desirable, especially when taking into account that there are less than half as many of them (around 500) as there are of the other more natural tones (around 1000). Whereas the rarity of the purple “night goddesses” at only 85 of them seems to more than compensate for their weird color.
Or perhaps people just prefer purple. Adults did in the Boulder school project.
Discarding the aliens
Things become a lot clearer when the red, green, blue, purple, golden and rainbow colored avatars are discarded, and we are left with the natural tones with approximately equivalent amounts:
And there you have it — the colors may not be arranged in a perfect spectrum, but it is close.
The real give-away is that three of the four most expensive categories have the word “light” in the description, and none contain the word “deep” (which I think we can agree is a synonym for “dark”). Similarly, three of the four least expensive categories contain the word “deep”, and none contain the word “light”.
Once again, I admit that this is not the most rigorous of studies. At the very least it should be repeated many times over the next few weeks.
I also think it is important to point out that World Of Women has clearly made an effort to have a fair spectrum of representation in their collection. And also that OpenSea is merely holding a mirror up for us to see what the biases of the community are.
However, when considering the above investigation, if I had to gamble on the outcome of a proper, full, and thorough investigation where would I place my bet?
Sadly, my money would be on the results I found being replicated.
Further research
There are plenty of opportunities for further analysis.
In the comments on Jason’s post, Demetrio Wazar-Santana asked if it would be possible to determine if the price is a reflection of the overall attitudes of the seller or the buyer.
There are other PFP NFT projects out there for which hypotheses could be proposed and that could be examined — World of Women is presumably popular with women, and I would guess that the female community of NFT enthusiasts is likely to be more racially diverse than the male community is. Is it therefore possible that, for example, the Meebit buying and selling community would show even more bias?
The nature of the bias could also be investigated. Is it a subconscious or even conscious racially motivated bias? Or is it a market phenomenon?
If people prefer to buy avatars that resemble them, and if, say, 60% of the buyers have a “light warm yellow” skin tone whereas only 10% have a “deep bronze skin” tone, but the number of each token type is about 1000, then mindless market forces could possibly account for the bias. Which itself could be a reflection of a racial wealth gap and corresponding differences in excess disposable income able to be spent on something most people would classify as frivolous, or at best a luxury.
And the final question is: what, if anything, can and should we do about it?
